PROJECT SUMMARY Sensorimotor transformations are mediated by premotor brain networks where individual neurons represent sensory, cognitive, and movement-related information. In the superior colliculus (SC), a central hub for producing visually-guided saccadic eye movements, many neurons emit a burst of action potentials both in response to a visual stimulus and when generating an eye movement command. These so-called visuomotor neurons project to the brainstem burst generator that produces the saccade. Thus, this downstream element is challenged to differentiate between the incoming ?visual? and ?motor? bursts. Multiple mechanisms have been proposed to account for movement generation. The ?fixed threshold? hypothesis posits that a saccade is produced once the firing rate of either an individual neuron or across a population crosses a threshold, which only happens during the motor burst. Existing data, however, indicate that a simple thresholding mechanism is likely not sufficient and requires consideration of other frameworks. The ?optimal subspace? hypothesis uses a dynamical systems approach to propose that a movement is generated when the population activity enters or resides within a particular region of state space. This implies that the state space representations of SC visual and motor bursts are dissociable. The ?temporal stability? hypothesis states that a movement is generated when bursting activity across a population of neurons preserves consistent temporal structure for a period of time. Indeed, the stability of SC population activity is reduced during a visual response (?unstable? temporal structure) and increased during an eye movement (?stable? temporal structure). We seek a framework that reconciles these models. Our central hypothesis is that SC population activity is decoded as a movement command when it both exhibits high temporal structure and resides within an optimal subspace. Our specific aim is to employ a closed-loop brain-computer interface in which monkeys are trained to control an auditory cursor by volitionally modulating the activity pattern across multiple SC neurons to lie within a visual or motor subspace and to be temporally stable or unstable. We will first test the optimal subspace and temporal stability frameworks individually before pitting the two against each other in a 2x2 design. Examining the trials in which an eye movement is observed will reveal the patterns used by population activity to represent a movement command. We predict that the animals will be able to modulate population activity along both visual-motor subspace and stable-unstable dimensions, but that the likelihood of movement generation will be the highest when the population activity is both stable and in the optimal subspace.